In this paper, we propose a new method for reducing the number of distance calculations in the LBG (Linde, Buzo, Gray) algorithm, which is widely used method to construct a codebook in vector quantization of speech recognition system. The proposed algorithm can reduce the distance calculation between input vector and codeword by utilizing the observation that codewords are quickly stabilized as the number of iteration increases. From the simulation results, it is shown that we can reduce the running times over $43.77\%$ on average in comparison with current LBG algorithm without sacrificing the performance of codebook.

A new algorithm for speech signal segmentation is proposed. This algorithm is based on finding successive similar frames belonging to a segment and represents it by an average spectrum. The speech signal is a slowly time varying signal in the sense that, when examined over a sufficiently short period of time (between 10 and 100 ms), its characteristics are fairly stationary. Generally this approach is based on finding these fairly stationary periods. Advantages of the. algorithm are accurate border decision of segments and simple computation. The automatic segmentations using frame averaging show as much as $82.20\%$ coincided with manually verified segmentation of CMU ARCTIC corpus within time range 16 ms. More than $90\%$ segment boundaries are coincided within a range of 32 ms. Also it can be combined with many types of automatic segmentations (HMM based, acoustic cues or feature based etc.).

The purpose of this study is to measure the acoustic properties of Tofu and Acorn Curd (Dotori Muk), which are possibly used as tissue mimicking materials (TMMs). Due to its availability and low cost, Tofu was suggested as a TMM by several researchers who measured only sound speed and attenuation. The acoustic properties of Tofu and Muk including the backscattering coefficient were measured in this paper. Sound speed was measured by the time shift in a pulse echo setup. Attenuation coefficients and backscattering coefficients were measured by a broadband method using both 5 MHz and 10 MHz transducers in the frequency domain. The measured acoustic properties of both Tofu and Muk are observed to be similar to those of biological tissues such as beef liver or beef heart.

The present study aims to investigate the effects of speaker location on the speech intelligibility in a classroom. In order to this, acoustic measurements were undertaken in a classroom with three different sound source locations such as center of front wall (FC), both sides of front wall (FS) and the center of ceiling (CC). SPL, RT, $D_{50}$, RASTI were measured in the 9 measurement points with same sound power level of sound source and MLS was used as the sound source signal. Also, subjective listening tests were carried out using Korean language listening materials which were recorded in an anechoic chamber. The recorded syllables were replayed and recorded again in the classroom with same sound source at three different locations and listening tests were undertaken to 20 respondents who were asked to write the correct syllables which were recorded in the classroom. The results show that higher sound intelligibility ($D_{50}$ of $47\%$, RASTI of 0.56) was obtained when sound source was located at the FS. The results also show that high sound intelligibility was obtained at the area nearby walls.

We evaluate the performance of emotion recognition via speech signals when a plain speaker talks to an entertainment robot. For each frame of a speech utterance, we extract the frame-based features: pitch, energy, formant, band energies, mel frequency cepstral coefficients (MFCCs), and velocity/acceleration of pitch and MFCCs. For discriminative classifiers, a fixed-length utterance-based feature vector is computed from the statistics of the frame-based features. Using a speaker-independent database, we evaluate the performance of two promising classifiers: support vector machine (SVM) and hidden Markov model (HMM). For angry/bored/happy/neutral/sad emotion classification, the SVM and HMM classifiers yield $42.3\%\;and\;40.8\%$ accuracy, respectively. We show that the accuracy is significant compared to the performance by foreign human listeners.

In this paper, a new data-aided joint phase and frequency estimator, which has very low computational complexity, is proposed and its variances of phase and frequency estimates are derived. To estimate the phase and frequency offset, first of all, the overall observation interval is divided into same length sub-intervals, and then phase estimates are independently computed based on symbols of the each sub-intervals. To be continue the sequence of computed phase estimates, proper integer multiples of $2{\pi}$ are added to (or subtracted from) the computed phase estimates, which is called linearized phase estimate. The phase offset of the proposed joint estimator is estimated by averaging the linearized phase estimates and the frequency offset by averaging the differences between consecutive linearized phase estimates. The variance of the proposed phase offset estimate is same to MCRB of phase if there is no frequency offset, but it is smaller than MCRB of phase if there is frequency offset. However, the variance of the proposed frequency offset estimate is bigger by at least 0.5 dB than MCRB of frequency with the same observation interval.